StopWatch
A research paper on object goal navigation that proposes making the goal grounding process explicit in order to improve navigation success rates, and that corrects annotation defects in existing benchmarks.
Timeline (last 90 days)
StopWatch finds that object grounding failure is the main bottleneck constraining the success rate of object-goal navigation.
StopWatch achieves state-of-the-art results on six benchmarks across three scene sets through explicit object grounding (locating the target, judging visibility, and estimating distance).
StopWatch corrects annotation flaws in existing benchmarks, making evaluation more trustworthy.
On the clean benchmark, when ground-truth object grounding information is provided, StopWatch's policy performance can nearly match the expert policy that generated its demonstration data.
The effectiveness of StopWatch shows that when a task has clear structure, explicitly modeling that structure improves generalization and sample efficiency more than simply scaling up model size.
StopWatch's method does not introduce complex network architectures or large-scale additional training.
All Facts (6)
StopWatch finds that object grounding failure is the main bottleneck constraining the success rate of object-goal navigation.
50%UnverifiedStopWatch achieves state-of-the-art results on six benchmarks across three scene sets through explicit object grounding (locating the target, judging visibility, and estimating distance).
50%UnverifiedStopWatch corrects annotation flaws in existing benchmarks, making evaluation more trustworthy.
50%UnverifiedOn the clean benchmark, when ground-truth object grounding information is provided, StopWatch's policy performance can nearly match the expert policy that generated its demonstration data.
50%UnverifiedThe effectiveness of StopWatch shows that when a task has clear structure, explicitly modeling that structure improves generalization and sample efficiency more than simply scaling up model size.
50%UnverifiedStopWatch's method does not introduce complex network architectures or large-scale additional training.
50%